A new artificial intelligence-powered optical coherence tomography (AI-OCT) system could significantly reduce unnecessary referrals for diabetic macular oedema (DME) while maintaining perfect sensitivity, according to a new study.
Researchers found that adding the AI-OCT system to an existing diabetic retinopathy screening pathway reduced false-positive referrals by almost two-thirds, potentially easing pressure on specialist ophthalmology services and improving patient care.
They said that while diabetic retinopathy screening programs commonly use retinal fundus photography to identify patients who may need further assessment, the approach can generate a high number of false-positive referrals for DME, resulting in unnecessary specialist appointments.
The study, published online in JAMA on 15 June 2026, evaluated an AI-based OCT system designed to act as a secondary screening tool after initial fundus photography.
The researchers said OCT imaging provides detailed cross-sectional views of the retina and is considered the gold standard for diagnosing DME but access to OCT interpretation can be limited by workforce and resource constraints.
The research was conducted in two stages. A prospective “silent mode” validation study involving 603 patients assessed the diagnostic performance of the AI system without influencing clinical decisions. The system demonstrated a sensitivity of 98.8% and specificity of 90.7% for detecting DME.
Researchers then conducted a multicentre, noninferiority randomised clinical trial involving 276 patients whose screening reports indicated suspected DME.
Participants were randomly assigned to standard care, in which all patients with suspected DME were referred for specialist assessment, or an intervention pathway that incorporated the AI-OCT system to determine whether referral was necessary.
No DME cases missed
The results showed a substantial reduction in false-positive referrals. The false-positive referral rate was 24.1% in the AI-OCT group compared with 69.1% in the standard-care group.
Importantly, the AI-assisted pathway maintained 100% sensitivity for DME referrals, meaning no patients with DME were missed.
Specificity improved dramatically, reaching 86.5% in the intervention group compared with 0% in the control group, where every patient flagged by fundus photography was automatically referred.
“No cases of DME occurred among non-referred participants in the intervention group,” the researchers said.
They said findings demonstrate that AI can be integrated safely into real-world clinical workflows rather than functioning solely as a diagnostic support tool in research settings.
The study was designed as a noninferiority trial, with investigators aiming to show that the AI-supported pathway would not perform worse than standard practice while reducing unnecessary referrals. The system comfortably met its’ prespecified noninferiority criterion and exceeded expectations by delivering a 45% reduction in false-positive referrals.
The AI-OCT platform incorporates automated image-quality assessment, DME detection and uncertainty flagging. During the validation phase, only 7.2% of scans were deemed ungradable, while 4.4% were classified as uncertain and could be directed for further review.
The researchers believe the technology could help diabetic retinopathy screening programs become more efficient as diabetes prevalence continues to rise globally.
They said the study provides “a practical framework for the real-world implementation of AI-enabled tools in ophthalmology and other clinical specialties”.
For optometrists, the findings highlight the growing role of AI-assisted imaging technologies in improving triage and referral pathways, potentially enabling earlier identification of sight-threatening disease while reducing unnecessary demands on specialist eye care services.
The researchers were from Hong Kong, London, China and Singapore.
